Bayesian Network Inference Modeling Identifies TRIB1 As a Novel Regulator of Cell

Author Manuscript Published OnlineFirst on January 13, 2017; DOI: 10.1158/0008-5472.CAN-16-0512 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Bayesian network inference modeling identifies TRIB1 as a novel regulator of cell

cycle progression and survival in cancer cells.

Rina Gendelman1, Heming Xing2, Olga K. Mirzoeva1, Preeti Sarde3, Christina Curtis4,

Heidi S. Feiler5, Paul McDonagh6, Joe W. Gray5, Iya Khalil7, W. Michael Korn1,8

Affiliations: 1Divisions of Gastroenterology and Hematology/Oncology, Department of Medicine, University of California, San Francisco, CA 2Novartis Institutes for BioMedical Research, Inc., Cambridge, MA 3Pharmacyclics, Sunnyvale, CA 4Departments of Medicine and Genetics, School of Medicine, Stanford University, Stanford, CA 5Oregon Health and Sciences University, Portland, OR 6Alexion Pharmaceuticals, Lexington, MA 7GNS Healthcare, Cambridge, MA 8Helen Diller Family Comprehensive Cancer Center, University of California San Francisco, CA

Running title: TRIB1 regulates cell cycle and survival. Key words: Molecular network biology, MEK, cell cycle regulation, apoptosis, TRIB1, NFκB Conflict of Interest: Ownership: GNS (H.X, P.M., I.K.) Financial Support: National Institutes of Health, National Cancer Institute grant U54 CA 112970 (J.W. Gray and W.M. Korn).

______

5Corresponding author: W. Michael Korn, M.D. Professor of Medicine UCSF Divisions of Gastroenterology and Hematology/Oncology, Department of Medicine Helen Diller Family Comprehensive Cancer Center 2340 Sutter St., Box 1387 San Francisco, CA 94115, USA Phone office: (415) 502 2844 Email: [email protected]

Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 13, 2017; DOI: 10.1158/0008-5472.CAN-16-0512 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Abstract

Molecular networks governing responses to targeted therapies in cancer cells are complex

dynamic systems that demonstrate non-intuitive behaviors. We applied a novel

computational strategy to infer probabilistic causal relationships between network

components based on gene expression. We constructed model comprised of an ensemble

of networks using multi-dimensional data from cell line models of cell cycle arrest caused

by inhibition of MEK1/2. Through simulation of reverse-engineered Bayesian network

model, we generated predictions of G1-S transition. The model identified known

components of the cell cycle machinery, such as CCND1, CCNE2, and CDC25A, as well

as revealed novel regulators of G1-S transition, IER2, TRIB1, TRIM27. Experimental

validation of model predictions confirmed 10 out 12 predicted genes to have a role in G1-

S progression. Further analysis showed that TRIB1 regulated the cyclin D1 promoter via

NFκB and AP-1 sites and sensitized cells to TNF-Related Apoptosis-Inducing Ligand

(TRAIL)-induced apoptosis. In clinical specimens of breast cancer, TRIB1 levels

correlated with expression of NFκB and its target genes (IL8, CSF2), and TRIB1 copy

number and expression were predictive of clinical outcome. Together our results establish

a critical role of TRIB1 in cell cycle and survival that is mediated via the modulation of

NFkB signaling.

The RAF-MEK-ERK signaling network is a key module of cellular signal integration and

transcriptional regulation. Extracellular stimuli (i.e. growth factors, cytokines, chemokines,

integrins, etc.) lead to activation of this network bringing on changes in global gene

expression and cellular outcomes, such as cell growth, proliferation, migration, and

survival. Much research has been focused on the role of Mitogen-Activated Protein Kinase

(MAPK) pathway in cancer progression and multiple MEK1/2 inhibitors have been

developed and tested in the clinic with limited success as single agents (1,2). It has been

well documented by others and our previous work that in cell line models of breast cancer,

inhibition of MEK results in cell cycle arrest but not cell death (3,4). We have previously

reported activation of a feedback-loop leading to Epidermal Growth Factor receptor

(EGFR)-dependent upregulation of Phosphatidyl Inositol 3-Kinase (PI3K) signaling in

response to MEK1/2 inhibition (3). This feedback mechanism contributes to resistance to

MEK inhibitor-induced apoptosis in these cells.

The MEK-PI3K feedback mechanism highlights the complexities of signal

transduction networks that govern non-intuitive responses of cellular systems to

pharmacological inhibitors. Reliable network models would be important to make

predictions about network responses to signal transduction inhibitors. This would allow to

efficiently identify key network nodes that could be targeted therapeutically, in particular

as part of combination therapies. Various modeling technologies have emerged that

capture network complexities in different ways (5). Among those, Bayesian network

inference models are particularly promising since they are not only capable of predicting

complex network structures but also suggest directionality of interactions between

network nodes, which leads to in silico hypotheses generation (6,7).

In order to build a model of transcriptional and cellular responses to MEK

inhibition, we assessed time-dependent changes in mRNA expression profiles and cell

cycle distribution following MEK inhibition in breast cancer cells. Using a novel Bayesian

network inference computational engine (6), ensembles of networks were calculated that

revealed novel MEK-dependent regulators of the cell cycle and suggested so far unknown

mechanisms of pathway cross-talk with the NFκB network. These model predictions were

experimentally validated in cell culture models and demonstrate a role of one of the MEK-

regulated genes, TRIB1, in mediating cross-talk with the NFκB network. Further analyses

in a large cohort of human breast cancers provided evidence for the existence of a TRIB1-

NFκB network in tumors and revealed TRIB1 as a predictor of breast cancer-free survival.

Materials and Methods.

Reagents. The following reagents were used: U0126 (Promega), epidermal growth factor

(EGF; Millipore), mimosine (Sigma), rhTRAIL (Millipore), rhTNFα (Life Tech),

TriplePrep Kit (GE-Healthcare). ON-TARGET plus SMARTpools siRNAs, NC (non-

coding negative control oligos) and individual oligos constituting the pools were

purchased from Dharmacon. RNAiMax and Lipofectamine LTX transfection reagents

were from Invitrogen. Antibodies: αR-TRIB1 (Millipore), αR-CCND1, αM-CCNA2, αR-

CDC25A, αR-IER2, αR-pCDK2, αR-pIKKa (Santa Cruz Biotechnology), αM-FLIP

(Enzo Life Sciences), αM-BID (BD Biosciences), αM-DR5 (R&D Systems), all other

antibodies were from Cell Signaling Technologies.

DNA Constructs. The TRIB1-EGFP construct was a generous gift from Dr. Kiss-Toth.

The cyclin D1 promoter-containing construct pD1luc WT and mutant promoter constructs

D1-κB1/2m, harboring two point mutations in the D1-κB1 (CGCGACCCCC) and the D1-

κB2 (CGCGAGTTTT) binding site (introduced point mutations are underlined), were a

gift from Dr. Hinz (Max-Delbrück-Center for Molecular Medicine, Berlin, Germany).

AP-1 mutant (AP1m) and EtsA/EtsB double mutant (EtsA/Bm) CCND1 promoter

constructs were generated by site-directed mutagenesis of pD1LucWT construct. NFkB-

Luc, pMetLuc-C vector, SEAP vector reporter constructs were from Clontech.

NFκB promoter reporter assay. Cells were co-transfected with 500ng of NFκB -Luc and

250ng of SEAP transfection-control vectors for 18 hours then treated with 10ng/ml TNFα.

Activation of NFκB promoter was assayed using Ready-To-Glow™ Dual Secreted

Reporter Assay system (Clontech) according to the manufacturers instructions 24h post

TNFα treatment.

Cell culture. MDA-MB-231, SUM149PT, MDA-MB-436, MDA-MB-468 triple-negative

basal breast cancer cell lines were obtained from ATCC (Manassas, VA) and

authenticated before experimental work began by single tandem repeat analysis at 15

different gene loci and amelogenin (Genetica, Burlington, NC). Cell line authentication

was performed by Dr. Gray and colleagues. Information about cell culture conditions as

well as the source, authentication, clinical, and pathological features of tumors used to

derive the breast cancer cell lines used in this study was described in detail previously (8).

siRNA treatment and Synchronization. The cells were transfected with 50nM specified

siRNA pools or non-coding control, according to the manufacturer’s instructions using

RNAiMax (Invitrogen) transfection reagent. Four hours post-transfection, the medium was

replaced to the one containing 0.4mM mimosine for 16h. Cells were released from

blocking and allowed to progress through the cell cycle for 12h, after which cells were re-

blocked with mimosine for 12h. Cells were collected at 10h post-mimosine release for cell

cycle analysis. Cell lysates for RNA and protein extraction were collected at 0, and 10h

post release from mimosine block.

Cell cycle, apoptosis analysis, and immunoblots. Cell cycle and apoptosis analysis were

performed by fluorescence-activated cell sorting (FACS) as well as standard immune blots

were generated as described before (3).

Real-Time Quantitative RT-PCR.

Total RNA was extracted from cells at 24h and 72h post-siRNA transfection using

RNeasy Micro kit (Qiagen). It was reverse-transcribed to cDNA and quantitative RT-PCR

analysis using the Taqman assay (ABI) was performed at Genome Analysis Core Facility

of Helen Diller Family Comprehensive Cancer Center, UCSF. PCR primers and TaqMan

probes for CCND1, TRIB1, IER2, CDKN2C, NUAK1, C14ORF133, CCNE2, TBK1,

EGR1, NPC1, SPRED2, KIAA0649, DR5, and YY1 were purchased from Applied

Biosystems. hGUS was used as a normalization control. The details of QPCR are

described in Supplementary methods.

Transcriptional assessment of MEK inhibition. We assessed the temporal changes in

gene expression profiles induced by EGF and UO126 in the MDAMB231 cell line by

RNA expression array technology as described in Supplementary Methods.

TaqMan Low density Array (TLDA). Quantitative PCRs were performed using custom-

made TLDA focused on 42 cell cycle related genes (Applied Biosystems; Supplementary

Table 3). Normalization of cDNA levels in the different wells was done using the GAPDH

and hGUS RNA as a reference. The amplification protocol and data analysis were carried

out as previously described (9). Initial data analysis was performed using Qbase-plus

software from Biogazelle (http://www.biogazelle.com/). Each sample set was analyzed

using the standardization procedure described previously (10).

Microarray RNA expression analaysis. Cell line and treatment preparation: 24 hours

prior to treatment, cells were transferred to low serum conditions (0.1% FBS), then they

were subjected to the following course of treatment. First, we pre-treated the cells with

U0126 (10μM) or DMSO (vehicle control) for 30 minutes (time -0.5h). At time 0h, the

cells were stimulated with EGF (10ng/ml). Cells were harvested at 8 timepoints: 0, 0.5, 1,

4, 8, 12, 24 and 48h after EGF stimulation (two biological replicates). Total RNA was

extracted using the RNeasy Mini kit (Qiagen). Microarray data were generated at the

LBNL Molecular Profiling Laboratory using a high-throughput, automated GeneChip®

HT System (Affymetrix). HT_HG-U133A array plates consisting of 96 arrays were used.

An Affymetrix liquid handling system (GCAS) was used to process the one-cycle IVT

target preparation, array plate hybridization setup, washing and staining. Plate scanning

was performed using a CCD-based high-throughput scanner (Affymetrix). Expression data

have been deposited in the NIH Gene Expression Omnibus

(http://www.ncbi.nlm.nih.gov/geo/); accession number: GSE61364.

Processing of microarray data, gene selection, and data frame construction. The

quality of the microarray data was assessed using the Simpleaffy package from

R/Bioconductor. Microarray data were then normalized using the Robust Multi-array

(RMA) method. A subset of genes was then selected to construct the gene regulatory

network based on analysis of covariance model to evaluate the significance of each gene

in explaining cell cycle progression. The 149 top-ranked genes were identified for

inclusion. This list was further supplemented by 20 known cell cycle genes, whose

expression was inhibited by treatment with U0126. To learn relationships between drug

treatment, mRNA expression profiles, and cell cycle distribution, gene expression

measurements at 0, 0.5, and 24 hours were matched with cell cycle distribution data at 0,

24 and 48 hours post-treatment.

Learning an ensemble of probabilistic models from data. An ensemble of Bayesian

network models including drug treatment, gene expression and end-point readouts was

learned from the data. Bayesian networks provide a convenient framework to represent

the global joint probability distribution of P(X1…….Xn : Q), where X1, … Xn are variables

in the model and Θ are parameters. As described previously (11), the first phase of

learning proceeds by considering all possible combinations of drug, transcripts, and

phenotypic data to obtain a collection of highly likely network fragments. Network

fragments representing quantitative relationships among all possible sets of two and three

measured variables were considered. Each fragment was assigned a Bayesian probabilistic

score indicative of how likely the given fragment was given the data, penalized for its

mathematical complexity. Here the Laplace approximation (11) was used to integrate

over posterior parameter distributions to compute the Bayesian scores for each fragment.

In the second phase of learning (6), the network fragments were combined to form an

ensemble of models. An ensemble is defined as a statistical sample from the distribution P

(Model|Data), the probability distribution over all models given the observed data. A

network’s score was computed as the sum over the scores of all network component

fragments. A parallel global optimization procedure employing simulated annealing (the

Metropolis-Hastings algorithm) was used to construct an ensemble of 1024 networks that

correctly sampled the probability distribution given the data (6,12).

Model interventional simulations and statistical analysis. Stochastic simulations of the

models were used to generate predictions on the distribution of a variable under different

conditions (11). Simulations were performed on the ensembles to predict the distribution

of model variables under different perturbations. Then the predicted distribution of

networks in response to the perturbation was compared to the predicted distribution under

baseline expression level of the perturbed transcript. A Student’s t-test was used to

compute the significance of a transcript’s influence on the phenotype and the transcripts

were ranked by their p-values.

Identification of drivers of cell cycle progression. To identify genes that were drivers of

cell cycle progression, a systematic perturbation procedure was employed. Genes in the

model were set to baseline as well as 10 fold below baseline, and cell cycle distribution

(%G1) was predicted. All genes were evaluated with this procedure and the significance of

each gene was evaluated via a Student’s t-test. The genes were then ranked based on their

p-values, which correspond to their likeliness to be causal drivers of cell cycle progression.

The p-value was derived based on 30 random samples from the posterior distribution

under each simulation condition.

Gene expression, copy number, and survival analysis of primary tumors. In order to

evaluate the association between TRIB1 related signaling and clinical characteristics, we

exploited a large, clinically annotated breast cancer cohort for which paired Illumina HT-

12 expression data and Affymetrix SNP 6.0 copy number data were available (13). Here

we report on a subset of 1980 cases to evaluate the correlation between TRIB1 and

NFκB/TRAIL signaling as well as clinical outcome associations. In particular, we

computed the Pearson correlation between a panel of 38 genes highlighted by functional

analyses of breast cancer cell lines and Bayesian network analysis, and reported to be

associated with NFκB/TRAIL signaling. Details of this analysis are provided in

Supplementary Methods.

Results

A Bayesian network inference model identifies genes involved in G1 arrest in

response to MEK inhibition. To assess effects of MEK inhibition on global gene

expression profiles and cell cycle distribution, we treated MDA-MB-231 breast cancer

cells with U0126, a small-molecule inhibitor of MEK1/2, (Fig. S1). To infer relationships

between drug treatment, mRNA expression profiles, and cell cycle distribution, gene

expression measurements at 0, 0.5, and 24 hours were matched with cell cycle distribution

data at 0, 24 and 48 hours post-treatment. Initial statistical analysis of the data identified

149 genes that were significantly correlated with cell cycle arrest following drug treatment

(Table S1). An additional 20 known cell cycle genes, whose expression levels were

significantly changed by U0126, were also included in the analysis (Table S1). These data

were used to reverse-engineer an ensemble of 1024 Bayesian networks (Fig. 1). The

inclusion of the 20 known cell cycle genes impacted by U0126 served as a positive control

for testing the models as we expected the model predictions to recover some of the known

biology. The ensemble of networks was reverse-engineered from the data in two steps

(6,7,14,15). First, a Bayesian score, which takes into account the uncertainty in the

parameters and penalizes for complexity, was computed for all possible two and three-

element sub-networks. Second, Metropolis Monte Carlo global optimization was used to

infer a statistical sample, or ensemble, of network structures from these sub-networks that

was consistent with the data (6). Variance in the ensemble reflected uncertainty in network

structure given the information available in the data (6). In this analysis we retained both

the network topology structures as well as the parameterization of the sub-networks in

order to generate quantitative predictions on the effect of gene interventions that account

for uncertainty in network topologies (6). Next, systematic in silico simulations of 10-fold

knockdown of each of the 169 transcripts were completed to predict their effect on G1-S

transition. This was accomplished by setting the value of a gene expression variable to

10% of its original value, and then propagating the effects of the intervention through the

network structure, and generating out-of-sample simulated values for the G1 fraction

under the knockdown condition (Fig. S2). Since all networks in the ensemble of models

are likely given the data, the entire distribution of predictions generated from the ensemble

of networks was analyzed for statistical differences between the knockdown and no-

knockdown simulation conditions using a paired t-test that samples from the distribution

30 times.

Using this approach, we identified 12 genes as candidate regulators of G1-S

transition, based on a significance level of p<0.01 (Table S1). These genes included

known regulators of G1-S transition (CCND1 (16,17), CCNE2 (18), CDC25A (19),

CDKN2C (20), NUAK1 (21) as well as novel candidates (TRIB1, IER2, C14ORF133,

TAF11, EBAG9, COQ9, and TRIM27).

Experimental validation of model predictions. To validate the model predictions, we

assessed cell cycle distribution in cells treated with pools of short-interfering RNAs

(siRNA) (Fig. 2a). We utilized pools of siRNA since they have been demonstrated to

result in greater phenotypic penetrance and reduced off-target effects compared to single

siRNAs (22). Cells were treated with pools of siRNAs directed against the 12 candidate

genes and 5 genes (NPC1, EGR1, SPRED2, BTG3, KIAA0649) that were predicted to have

no effect on G1-S phase transition as negative controls, as well as non-coding control

siRNAs (Table S2). Target knock-down was confirmed by mRNA analysis and protein

levels (where antibodies were available) (Fig. S3). Fluorescence-activated cell sorting

(FACS) analysis of cell cycle distribution confirmed the established role of CCND1,

CDC25a, and CCNE2 in G1 to S progression, and validated model predictions for novel

candidate regulators such as IER2, TRIB1, NUAK1, C14ORF133, TRIM27, TAF11, and

EBAG9 (Fig. 2a). However, we saw no effect on G1-S transition with siRNAs directed

against CDKN2C and COQ9, while knockdown of KIAA0649, a negative control,

resulted in G1 arrest. Effect of individual siRNA duplexes against TRIB1 on cell cycle

was also analyzed (Fig. S4). Overall, the model predicted the effect of interventions with

82% accuracy, 91% sensitivity, and 67% specificity. To understand the mechanism of G1

arrest, we analyzed the status of proteins known to promote G1-S transition (CCND1,

CCNA2, CCNE2, phospho-CDK2, ppRB), as well as inhibitors (p27, p18) (Fig. 2b,c). We

found that knock down of genes leading to G1 arrest resulted in at least a 2-fold increase

in protein levels of p27, p18 or both, with the exception of TRIM27 and KIAA0649.

Simultaneously, knock down of these genes resulted in inhibition of at least three

promoters of G1-S transition (Fig. 2c). Analysis of mRNA expression of 46 genes

associated with G1-S progression showed distinct patterns of gene expression after siRNA

treatment that confirmed the protein findings for CCND1, CCNA2, and CCNE2 (Fig. S5).

TRIB1 regulates CCND1 expression via kappaB and AP1 sites. TRIB1 was predicted

by the model to regulate G1-S phase transition with the highest significance score

(p=2.49E-10). Its siRNA-mediated knockdown resulted in inhibition of cyclin D1 RNA and

protein levels (Fig. 2b, Fig. S5). To investigate the mechanism of TRIB1 knockdown

effect on CCND1 expression we evaluated its effect on the activity of CCND1 promoter

using constructs containing mutations in AP1, NFκB, and Ets transcription factor-binding

sites (Fig. 3a-b). In cells treated with NC siRNA, inactivation of either both NFκB (D1-

κB1/2M) sites or AP1 site led to inhibition of CCND1 promoter activity, while

inactivation of the two Ets sites had no effect. TRIB1 knockdown resulted in 2-fold

downregulation of wild type CCND1 (D1-WT) promoter activity with further reduction in

activity of D1-κB1/2M and AP1 promoter constructs, indicating that TRIB1 signals

upstream of both ERK1/2 and NFκB (Fig. 3b).

TRIB1 mediates NFkB-responsive promoter activity and expression of CXCL1,

CSF2, and IL8. To confirm the role of TRIB1 in regulation of NFκB activity we

analyzed the effect of TRIB1 knockdown and overexpression on an NFκB-responsive

promoter construct. This analysis showed that TRIB1 knockdown inhibited the activity of

an NFκB-responsive promoter, while TRIB1 overexpression led to a 3-fold increase in

promoter activity (Fig. 3c, Fig. S6). The expression of siRNA-resistant TRIB1 rescued the

phenotype induced by siRNA-mediated TRIB1 knockdown (Fig. S7). Furthermore,

evaluation of the expression of NFκB target genes involved in metastasis formation (CSF2,

CXCL1) and tumor inflammation (IL8) showed that inhibition of TRIB1 with siRNA

resulted in downregulation of these genes on RNA and protein levels (Fig. 3d, Fig. S8).

Activation of NFκB is mediated by inactivation of its inhibitors IκBα, p100 and p105 (23)

by several upstream kinases, including IKKα. We reasoned that TRIB1 may act upstream

of NFκB. TRIB1 knockdown inhibited phosphorylation and degradation of IκBα as well

as production of p50 from its precursor p105 (Fig.3e, Fig. S9). Additionally, we found that

phosphorylation of IKKα/β (Ser176/180), of IKKα (Thr 23) as well as expression of p100

and its processing with generation of p52 were significantly inhibited by downregulation

of TRIB1, possibly indicating regulation of p100 expression by TRIB1 (Fig. 3e, Fig S9).

These findings provide mechanistic evidence for regulation of NFκB by TRIB1. No

changes in RalA phosphorylation at serine 536 were observed. While IKKα and AKT

have been shown to phosphorylate RelA, multiple additional kinases are known to

phosphorylate RelA at the same site and might maintain phosphorylation following knock-

down of TRIB1 (24,25).

TRIB1 regulates activity of the PI3K-AKT pathway. One of the potential mechanisms

of the cytostatic rather then cytotoxic effect of MEK inhibition is activation of a feedback

loop leading to upregulation of AKT signaling (3). In contrast, TRIB1 knockdown leads to

simultaneous inhibition of phosphorylation of ERK1/2 and AKT1 on both T308 and S473

(Fig. 4a). Assessment of downstream targets of AKT kinase revealed that TRIB1

knockdown led to inhibition of phosphorylation of BCL2, GSK3α/β, and FOXO1A/3A

(Fig. 4a). Additionally, we observed inhibition of phosphorylation of IKKα-T23, an AKT-

specific site, providing further evidence for a putative mechanism of NFκB regulation by

TRIB1 through its role as a mediator of PI3-kinase-AKT signaling (Fig. 3e).

TRIB1 inhibition upregulates DR5 levels and sensitizes breast cancer cells to TRAIL-

induced apoptosis. We postulated that inhibition of anti-apoptotic and induction of pro-

apoptotic signals results in increased cell death upon TRIB1 knockdown alone or in

combination with the death receptor agonist TRAIL. Treatment of cells with TRIB1

siRNA alone resulted in an increase in apoptosis in all cell lines tested (Fig. 4b). We also

observed a 2-3 fold increase in apoptosis in cells treated with TRIB1 siRNA in

combination with TRAIL treatment (Fig. 4c). These observations were in accord with the

induction of PARP cleavage (Fig. 4d). TRAIL-induced apoptosis occurs through an

extrinsic mechanism of direct activation of executioner caspases by receptor-activated

caspase-8 (26). In addition, apoptosis can be amplified via recruitment of the intrinsic

mitochondrial pathway. Activated caspase-8 cleaves pro-apoptotic protein BID, leading

to its translocation to the mitochondria and resulting in mitochondrial apoptosis activation

(26). TRIB1 knockdown resulted in increased caspase-8 cleavage (Fig. S10), as well as

decreased levels of FLIP and BID, indicating increased cleavage of these proteins and the

involvement of both pathways (Fig. 4d). To explain how TRIB1 inhibition leads to up-

regulation of TRAIL-induced apoptosis we assayed both mRNA and protein levels of

TRAIL Receptor 2 (DR5). DR5 expression is negatively regulated by NFκB via induction

of the transcriptional inhibitor YY1 (27). Here we show that knockdown of TRIB1 leads

to significant upregulation of both, DR5 mRNA and protein levels, as well as

downregulation of YY1 mRNA (Fig. 4e,f; Fig. S11). These data suggest that TRIB1

knockdown sensitizes cells to TRAIL-induced apoptosis by inhibition of NFκB signaling

and upregulation of DR5, as well as by inhibition of AKT pro-survival signaling.

Association between TRIB1/TRAIL/NFkB signaling components in human breast

cancer. To establish the significance of TRIB1/TRAIL/NFkB signaling in the context of

human disease and to perform exploratory tests of clinical outcome associations, we

exploited a large cohort (n=1980) of primary breast tumors, with long-term clinical

follow-up and which have been profiled at the genomic and transcriptomic levels (13).

Given our finding that TRIB1 mediates cyclin D1 expression via the regulation of NFκB

and AP1, with TRIB1 knockdown resulting in the downregulation of NFκB targets in

vitro, we first sought to examine these relationships in the human system. In agreement

with our earlier findings, TRIB1 and NFKB1(p105) expression levels were significantly

correlated (ρ=0.102, p<0.0001), and TRIB1 was correlated with the expression of the

NFκB target genes IL8 (ρ=0.05, p<0.05) and CSF2 (ρ=0.05, p<0.005). We also

comprehensively examined the relationships between TRIB1, TRAIL, and NFκB

signaling using this tumor cohort to compare the correlations between a panel of 38 genes

believed to be associated with these pathways based on the literature (Fig. 5a). In addition

to those listed above, we found that the following gene sets were significantly (p<0.05)

positively correlated with TRIB1: proliferation associated genes (CCND1, TOP2A, MCM2,

MKI67), the anti-apoptotic genes (BCL2L1, BCL2L11, CFLAR), the pro-apoptotic genes

(BIK, HRK), components of the death receptor complex (TRAF2, TRAF6), as well as

TNFRSF10A/DR4 and IL6. A small number of genes in each of these categories were

negatively correlated with TRIB1, including PCNA (proliferation), BCL2LA1 (anti-

apoptotic) and BAD (pro-apoptotic), TRADD and MPRIP (death receptor complex), and

CASP3 (caspase activity) (Fig. 5a). These findings are in agreement with our observation

that TRIB1 knockdown sensitizes cells to TRAIL-induced apoptosis, resulting in

increased cleavage of both BID and CFLAR in breast cancer cells and supports the role of

TRIB1 in cell cycle progression and survival.

Association between TRIB1/TRAIL/NFκB associated gene expression profiles and

TRIB1 copy number with clinical outcome. We next sought to explore the correlation

between TRIB1/TRAIL/NFkB associated genes and patient survival in a human breast

tumor cohort. A comparison of Kaplan-Meier curves for TRIB1 copy number amplified

versus neutral cases suggests an association with both breast cancer specific (BCSS) (log-

rank p-value <0.01) and overall survival (OS) (log-rank p-value <0.01) (Fig. 5b, Table S4).

In order to visualize the association between TRIB1 or other single gene expression

models and outcome in an exploratory analysis, continuous expression levels were

stratified into tertiles, and the predicted survival curves for individuals in the upper

(66.66%) and lower (33.33%) groups were compared using Kaplan-Meier plots and the

log-rank test. These comparisons suggest an association between both TRIB1 and IL8

expression and outcome, as assessed by both BCSS (log-rank p-value <0.01) and OS (log-

rank p-value <0.05) (Fig. S12a-d).

To investigate these relationships further, while adjusting for clinically relevant

covariates, we examined survival as a function of continuous gene expression data for

single and multiple gene models based on the panel of 38 TRIB1/TRAIL/NFkB associated

genes using Cox regression. In particular, multivariable analysis was performed by

building Cox proportional hazards models, which included the most relevant clinical

variables as covariates, namely, age, number of lymph nodes positive, tumor size, and

grade, stratified by ER status and tumor bank in order to test for differences in BCSS and

OS associated with these genes, as described in the Methods section. After adjusting for

these variables, only BIK, MPRIP, IL8, and TOP2A were significant (Wald test, p<0.05) in

the full cohort, where BCSS was the outcome (Fig. S13, Table S5). For OS BIK, MPRIP,

and TOP2A were again significant, as were HRK and CASP3, whereas IL8 was not

associated with this outcome (Fig. S14, Table S6).

Discussion

MEK-dependent signaling is a highly complex process as highlighted by our previously

described finding of an EGFR-dependent feedback loop resulting in activation of the PI3

kinase pathway following MEK inhibition (3). Studies broadly assessing protein

expression changes following MEK inhibitor treatment corroborate these findings (28).

Among the best-documented cellular consequences of MEK inhibition is induction of cell

cycle arrest, predominantly at the G1-S phase transition (3,29,30). This cytostatic effect is

mediated by a variety of mechanisms (reviewed in (31)) including repression of cyclin D1

(32). The work presented here sheds new light onto the intermediary steps downstream of

MEK and upstream of CCND1. We utilized an advanced Bayesian network inference

engine to deepen our understanding of molecular networks driving responses of cancer

cells to MEK inhibition.

Bayesian network inference models allow for reconstruction of molecular networks and

for making causal predictions about the relationships of individual network components.

While the Bayesian approach has been applied to the reconstruction of small signal

transduction networks before (33,34), our study involves automated construction of large

ensembles of networks and extensive experimental validation. Our model predictions

correctly identify the established role of CCND1, CCNE2, CDC25a, and ARK5/NUAK1

in regulation of G1-S progression. Moreover, we identified and experimentally validated

TRIB1, IER2, C14ORF133, TAF11, EBAG9, and TRIM27 as novel regulators of G1-S

transition. The high rate of validated predictions underscores the power of simulatable

Bayesian models.

In the model, TRIB1 was the highest ranked gene (p-val=2.49E-10) in our list of genes

predicted to affect cell cycle distribution. During Drosophila development, Tribbles has

been demonstrated to regulate G2-M transition during anlage formation through a cdc25-

dependent mechanism (35,36). TRIB1 has recently been shown to control differentiation

of M2-like macrophages in C/EBPα-dependent manner (37) and it has been implicated in

proliferation and chemotaxis of vascular smooth muscle cells via regulation of MAPK

activity (38). Most importantly in the context of the current study, TRIB1 and TRIB2 have

been demonstrated to be sufficient to induce leukemia in mice when overexpressed (39).

However, the role of TRIB1 in regulation of the mammalian cell cycle remains unknown.

In this study, we present evidence that TRIB1 is involved in regulation of G1-S

progression by modulating CCND1 expression in human breast cancer cells. CCND1 is

highly overexpressed and a pivotal driver of G1-S progression in these cells (40). The

CCND1 promoter contains binding sites for multiple transcription factors including four

Ets, two NFκB, and one AP1 site (41-43). We found that TRIB1 regulates NFκB and

AP1-dependent transactivation of the CCND1 gene. These findings support our

observation that TRIB1 knockdown leads to downregulation of CCND1 protein and RNA

levels. Indeed, TRIB1 was found to act upstream of the MAPKK SEK1 in C. elegance

(44) and it has been shown to physically interact with MEK1 and p65 subunit of NFκB in

mammalian cells and to promote induction of pro-inflammatory cytokines in lipocytes

(45). Moreover, NFκB has been demonstrated to enhance AP-1 dependent transactivation

in lymphocytes (46). Thus, our data illuminate a role for TRIB1 in regulating MAPK-

dependent signals, and, in a broader sense, in oncogenesis and inflammation. Interestingly,

our data demonstrate that TRIB1 RNA expression might be regulated through the MAPK

pathway, suggesting a feed-forward mechanism by which signals through the MAPK

pathway increase TRIB1 expression which, in turn, further stimulates signaling through

the pathway. This possibility is subject of ongoing investigation. Other TRIB-family

members, in particular TRIB3, have been shown to interact with MAPK and Notch

signaling in breast cancer, highlight these pseudo-kinases as modulators of key oncogenic

signaling pathways (47).

Recent studies revealed upregulation of TRIB1 during acute and chronic inflammation in

white adipose tissue of mice (48). In agreement with these findings, we observed

inhibition of CXCL1, CSF2 and IL8 expression in response to treatment with TRIB1

siRNA, as well as upregulation of these genes in cells overexpressing TRIB1 construct. It

remains to be determined whether TRIB1 serves as a nuclear co-activator of p65 subunit

in mammary epithelial cells. However, we found that knock-down of TRIB1 leads to

inhibition of pIKKα, pIκBα, and subsequent pIκBα degradation. These findings point to a

cytoplasmic role of TRIB1 in regulation of NFkB-dependent transcription.

Phosphorylation of IKKα on Tyrosine-23 has been shown to be strongly dependent on

AKT (49). In addition, we observed inhibition of multiple components of the PI3K

pathway in response to knock-down of TRIB1, providing evidence that TRIB1 influences

signal transduction of the PI3K-NFκB pathways beyond the nucleus. TRIB1 functions, at

least in part, as a scaffold protein (50) suggesting that TRIB1 might mediate protein-

protein interactions of key regulators of the PI3K and NFκB pathways resulting in their

activation. This possibility is currently under investigation.

Regulation of apoptosis represents one of the main functions of the NFκB pathway. In

solid tumors, the pathway predominantly suppresses apoptosis induction through a

multitude of mechanisms (reviewed in: (51)) and has been shown to be a major

contributor to resistance to chemotherapeutic agents (52). Suppression of death-receptor

signaling by NFkB has been implicated as one possible mechanism (53). Our work

demonstrates that knock-down of TRIB1 induces death receptor signaling by down-

regulation of DR5 transcriptional repressor, YY1, resulting in significant increase of DR5

levels. Regulation of YY1 by NFκB is well-documented (54). Despite promising

preclinical results, initial clinical experience with death receptor agonists for solid tumor

indications have been disappointing (55). Based on our findings, we speculate that high

expression levels of TRIB1 contribute to resistance to this class of drugs and that

inhibitors of TRIB1 activity could overcome therapeutic resistance.

Discovery of involvement of TRIB1 in regulating proliferation, apoptosis, and cytokine

production suggests that high levels of TRIB1 expression might result in a clinically more

aggressive tumor phenotype. In agreement with this hypothesis we found that increased

DNA copy numbers of the TRIB1 gene as well as high level of TRIB1 mRNA expression

are associated with poor breast cancer survival in a large cohort of breast cancers (13).

TRIB1 is located on chromosome 8q24 in immediate vicinity of the oncogene cMyc,

which is often amplified in cancer. Furthermore, we found TRIB1 expression to be

associated with elements of the NFκB network (e.g. NFKB1, IL8) that we had identified to

be down-stream of TRIB1 in our in vitro studies. Based on the association of TRIB1 with

a key survival endpoint in breast cancer and its biological functions, we are currently

pursuing studies to evaluate its potential as a target for therapeutic intervention.

In summary, our study highlights that linear signal transduction cascades, such as the

RAS-RAF-MEK-ERK module, are embedded in complex molecular networks. We show

that the combination of Bayesian network inference models, in silico simulation, and in

vitro validation represents a powerful new approach for discerning non-intuitive

relationships and generation of testable hypotheses about signaling networks from a large,

multi-dimensional data set. By applying this strategy, we identified TRIB1 as an important

mediator of key signal transduction pathways and a candidate therapeutic target in breast

cancer.

Acknowledgements We thank E. Kiss-Toth for generating TRIB1-EGFP vector and M.

Hinz for providing D1-WT and D1-kB1/2m constructs. We thank B. Church, K. Runge, R.

Miller, and D. Decaprio for development of REFS(TM) algorithms, software, and insights

on methodology utilized in this work. We thank P. Jensen and Boris Hayete for critically

reading the manuscript.

References

1. Wang D, Boerner SA, Winkler JD, LoRusso PM. Clinical experience of MEK inhibitors in cancer therapy. Biochim Biophys Acta 2007;1773(8):1248-55 doi S0167-4889(06)00369-7 [pii] 10.1016/j.bbamcr.2006.11.009. 2. Friday BB, Adjei AA. Advances in targeting the Ras/Raf/MEK/Erk mitogen- activated protein kinase cascade with MEK inhibitors for cancer therapy. Clin Cancer Res 2008;14(2):342-6 doi 10.1158/1078-0432.CCR-07-4790 14/2/342 [pii]. 3. Mirzoeva OK, Das D, Heiser LM, Bhattacharya S, Siwak D, Gendelman R, et al. Basal subtype and MAPK/ERK kinase (MEK)-phosphoinositide 3-kinase feedback signaling determine susceptibility of breast cancer cells to MEK inhibition. Cancer research 2009;69(2):565-72 doi 69/2/565 [pii] 10.1158/0008-5472.CAN-08-3389. 4. Turke AB, Song Y, Costa C, Cook R, Arteaga CL, Asara JM, et al. MEK inhibition leads to PI3K/AKT activation by relieving a negative feedback on ERBB receptors. Cancer research 2012;72(13):3228-37 doi 10.1158/0008- 5472.CAN-11-3747 0008-5472.CAN-11-3747 [pii]. 5. Janes KA, Lauffenburger DA. A biological approach to computational models of proteomic networks. Curr Opin Chem Biol 2006;10(1):73-80 doi S1367- 5931(05)00168-7 [pii] 10.1016/j.cbpa.2005.12.016. 6. Xing H, McDonagh PD, Bienkowska J, Cashorali T, Runge K, Miller RE, et al. Causal modeling using network ensemble simulations of genetic and gene expression data predicts genes involved in rheumatoid arthritis. PLoS Comput Biol 2011;7(3):e1001105 doi 10.1371/journal.pcbi.1001105. 7. Friedman N, Koller D. Being Bayesian About Network Structure. A Bayesian Approach to Structure Discovery in Bayesian Networks. Machine Learning 2003;50(1-2):95-125 doi 10.1023/a:1020249912095. 8. Neve RM, Chin K, Fridlyand J, Yeh J, Baehner FL, Fevr T, et al. A collection of breast cancer cell lines for the study of functionally distinct cancer subtypes. Cancer cell 2006;10(6):515-27. 9. Ermonval M, Petit D, Le Duc A, Kellermann O, Gallet PF. Glycosylation-related genes are variably expressed depending on the differentiation state of a bioaminergic neuronal cell line: implication for the cellular prion protein. Glycoconj J 2009;26(4):477-93 doi 10.1007/s10719-008-9198-5. 10. Willems E, Leyns L, Vandesompele J. Standardization of real-time PCR gene expression data from independent biological replicates. Anal Biochem 2008;379(1):127-9 doi 10.1016/j.ab.2008.04.036 S0003-2697(08)00264-9 [pii]. 11. Tierney L, Kadane JB. Accurate approximations for posterior moments and marginal densities. . J Ameri Stat Assoc 1986;81(393):82-6. 12. Ding Y, Lawrence CE. A statistical sampling algorithm for RNA secondary structure prediction. Nucleic Acids Res 2003;31(24):7280-301. 13. Curtis C, Shah SP, Chin SF, Turashvili G, Rueda OM, Dunning MJ, et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals

novel subgroups. Nature 2012;486(7403):346-52 doi 10.1038/nature10983 nature10983 [pii]. 14. Friedman N, Linial M, Nachman I, Pe'er D. Using Bayesian networks to analyze expression data. J Comput Biol 2000;7(3-4):601-20. 15. Sachs K, Perez O, Pe'er D, Lauffenburger DA, Nolan GP. Causal protein- signaling networks derived from multiparameter single-cell data. Science (New York, NY 2005;308(5721):523-9 doi 308/5721/523 [pii] 10.1126/science.1105809. 16. Baldin V, Lukas J, Marcote MJ, Pagano M, Draetta G. Cyclin D1 is a nuclear protein required for cell cycle progression in G1. Genes Dev 1993;7(5):812- 21. 17. Stacey DW. Cyclin D1 serves as a cell cycle regulatory switch in actively proliferating cells. Curr Opin Cell Biol 2003;15(2):158-63 doi S0955067403000085 [pii]. 18. Aleem E, Kiyokawa H, Kaldis P. Cdc2-cyclin E complexes regulate the G1/S phase transition. Nature cell biology 2005;7(8):831-6 doi ncb1284 [pii] 10.1038/ncb1284. 19. Jinno S, Suto K, Nagata A, Igarashi M, Kanaoka Y, Nojima H, et al. Cdc25A is a novel phosphatase functioning early in the cell cycle. EMBO J 1994;13(7):1549-56. 20. Latres E, Malumbres M, Sotillo R, Martin J, Ortega S, Martin-Caballero J, et al. Limited overlapping roles of P15(INK4b) and P18(INK4c) cell cycle inhibitors in proliferation and tumorigenesis. EMBO J 2000;19(13):3496-506 doi 10.1093/emboj/19.13.3496. 21. Humbert N, Navaratnam N, Augert A, Da Costa M, Martien S, Wang J, et al. Regulation of ploidy and senescence by the AMPK-related kinase NUAK1. EMBO J 2010;29(2):376-86 doi 10.1038/emboj.2009.342 emboj2009342 [pii]. 22. Parsons BD, Schindler A, Evans DH, Foley E. A direct phenotypic comparison of siRNA pools and multiple individual duplexes in a functional assay. PLoS One 2009;4(12):e8471 doi 10.1371/journal.pone.0008471. 23. Oeckinghaus A, Ghosh S. The NF-kappaB family of transcription factors and its regulation. Cold Spring Harb Perspect Biol 2009;1(4):a000034 doi 10.1101/cshperspect.a000034. 24. Bohuslav J, Chen LF, Kwon H, Mu Y, Greene WC. p53 induces NF-kappaB activation by an IkappaB kinase-independent mechanism involving phosphorylation of p65 by ribosomal S6 kinase 1. The Journal of biological chemistry 2004;279(25):26115-25 doi 10.1074/jbc.M313509200. 25. Ghosh S, Karin M. Missing pieces in the NF-kappaB puzzle. Cell 2002;109 Suppl:S81-96. 26. Johnstone RW, Frew AJ, Smyth MJ. The TRAIL apoptotic pathway in cancer onset, progression and therapy. Nat Rev Cancer 2008;8(10):782-98 doi nrc2465 [pii] 10.1038/nrc2465. 27. Martinez-Paniagua MA, Baritaki S, Huerta-Yepez S, Ortiz-Navarrete VF, Gonzalez-Bonilla C, Bonavida B, et al. Mcl-1 and YY1 inhibition and induction of DR5 by the BH3-mimetic Obatoclax (GX15-070) contribute in the

sensitization of B-NHL cells to TRAIL apoptosis. Cell Cycle 2011;10(16):2792- 805 doi 16952 [pii]. 28. Duncan JS, Whittle MC, Nakamura K, Abell AN, Midland AA, Zawistowski JS, et al. Dynamic reprogramming of the kinome in response to targeted MEK inhibition in triple-negative breast cancer. Cell 2012;149(2):307-21 doi 10.1016/j.cell.2012.02.053 S0092-8674(12)00350-9 [pii]. 29. Gysin S, Lee SH, Dean NM, McMahon M. Pharmacologic inhibition of RAF-- >MEK-->ERK signaling elicits pancreatic cancer cell cycle arrest through induced expression of p27Kip1. Cancer research 2005;65(11):4870-80 doi 65/11/4870 [pii] 10.1158/0008-5472.CAN-04-2848. 30. Sebolt-Leopold JS, Dudley DT, Herrera R, Van Becelaere K, Wiland A, Gowan RC, et al. Blockade of the MAP kinase pathway suppresses growth of colon tumors in vivo. Nat Med 1999;5(7):810-6. 31. Chambard JC, Lefloch R, Pouyssegur J, Lenormand P. ERK implication in cell cycle regulation. Biochim Biophys Acta 2007;1773(8):1299-310 doi S0167- 4889(06)00379-X [pii] 10.1016/j.bbamcr.2006.11.010. 32. Weber JD, Raben DM, Phillips PJ, Baldassare JJ. Sustained activation of extracellular-signal-regulated kinase 1 (ERK1) is required for the continued expression of cyclin D1 in G1 phase. The Biochemical journal 1997;326 ( Pt 1):61-8. 33. Hill SM, Lu Y, Molina J, Heiser LM, Spellman PT, Speed TP, et al. Bayesian inference of signaling network topology in a cancer cell line. Bioinformatics (Oxford, England) 2012;28(21):2804-10 doi 10.1093/bioinformatics/bts514 bts514 [pii]. 34. Froehlich H, Fellmann M, Sueltmann H, Poustka A, Beissbarth T. Large scale statistical inference of signaling pathways from RNAi and microarray data. BMC Bioinformatics 2007;8:386 doi 1471-2105-8-386 [pii] 10.1186/1471- 2105-8-386. 35. Seher TC, Leptin M. Tribbles, a cell-cycle brake that coordinates proliferation and morphogenesis during Drosophila gastrulation. Curr Biol 2000;10(11):623-9 doi S0960-9822(00)00502-9 [pii]. 36. Mata J, Curado S, Ephrussi A, Rorth P. Tribbles coordinates mitosis and morphogenesis in Drosophila by regulating string/CDC25 proteolysis. Cell 2000;101(5):511-22 doi S0092-8674(00)80861-2 [pii]. 37. Satoh T, Kidoya H, Naito H, Yamamoto M, Takemura N, Nakagawa K, et al. Critical role of Trib1 in differentiation of tissue-resident M2-like macrophages. Nature 2013;495(7442):524-8 doi 10.1038/nature11930 nature11930 [pii]. 38. Sung HY, Guan H, Czibula A, King AR, Eder K, Heath E, et al. Human tribbles-1 controls proliferation and chemotaxis of smooth muscle cells via MAPK signaling pathways. The Journal of biological chemistry 2007;282(25):18379- 87 doi M610792200 [pii] 10.1074/jbc.M610792200. 39. Dedhia PH, Keeshan K, Uljon S, Xu L, Vega ME, Shestova O, et al. Differential ability of Tribbles family members to promote degradation of C/EBPalpha and induce acute myelogenous leukemia. Blood 2010;116(8):1321-8 doi 10.1182/blood-2009-07-229450.

40. Hosokawa Y, Arnold A. Mechanism of cyclin D1 (CCND1, PRAD1) overexpression in human cancer cells: analysis of allele-specific expression. Genes Chromosomes Cancer 1998;22(1):66-71 doi 10.1002/(SICI)1098- 2264(199805)22:1<66::AID-GCC9>3.0.CO;2-5 [pii]. 41. Tetsu O, McCormick F. Beta-catenin regulates expression of cyclin D1 in colon carcinoma cells. Nature 1999;398(6726):422-6. 42. Hinz M, Krappmann D, Eichten A, Heder A, Scheidereit C, Strauss M. NF- kappaB function in growth control: regulation of cyclin D1 expression and G0/G1-to-S-phase transition. Mol Cell Biol 1999;19(4):2690-8. 43. Klein EA, Assoian RK. Transcriptional regulation of the cyclin D1 gene at a glance. J Cell Sci 2008;121(Pt 23):3853-7 doi 10.1242/jcs.039131. 44. Pujol N, Cypowyj S, Ziegler K, Millet A, Astrain A, Goncharov A, et al. Distinct innate immune responses to infection and wounding in the C. elegans epidermis. Curr Biol 2008;18(7):481-9 doi 10.1016/j.cub.2008.02.079 S0960-9822(08)00361-8 [pii]. 45. Kiss-Toth E, Bagstaff SM, Sung HY, Jozsa V, Dempsey C, Caunt JC, et al. Human tribbles, a protein family controlling mitogen-activated protein kinase cascades. The Journal of biological chemistry 2004;279(41):42703-8 doi 10.1074/jbc.M407732200 M407732200 [pii]. 46. Krappmann D, Wegener E, Sunami Y, Esen M, Thiel A, Mordmuller B, et al. The IkappaB kinase complex and NF-kappaB act as master regulators of lipopolysaccharide-induced gene expression and control subordinate activation of AP-1. Mol Cell Biol 2004;24(14):6488-500 doi 10.1128/MCB.24.14.6488-6500.2004 24/14/6488 [pii]. 47. Izrailit J, Berman HK, Datti A, Wrana JL, Reedijk M. High throughput kinase inhibitor screens reveal TRB3 and MAPK-ERK/TGFbeta pathways as fundamental Notch regulators in breast cancer. Proceedings of the National Academy of Sciences of the United States of America 2013;110(5):1714-9 doi 10.1073/pnas.1214014110. 48. Ostertag A, Jones A, Rose AJ, Liebert M, Kleinsorg S, Reimann A, et al. Control of adipose tissue inflammation through TRB1. Diabetes 2010;59(8):1991- 2000 doi 10.2337/db09-1537 db09-1537 [pii]. 49. Ozes ON, Mayo LD, Gustin JA, Pfeffer SR, Pfeffer LM, Donner DB. NF-kappaB activation by tumour necrosis factor requires the Akt serine-threonine kinase. Nature 1999;401(6748):82-5 doi 10.1038/43466. 50. Sung HY, Francis SE, Crossman DC, Kiss-Toth E. Regulation of expression and signalling modulator function of mammalian tribbles is cell-type specific. Immunol Lett 2006;104(1-2):171-7 doi 10.1016/j.imlet.2005.11.010. 51. Prasad S, Ravindran J, Aggarwal BB. NF-kappaB and cancer: how intimate is this relationship. Mol Cell Biochem 2010;336(1-2):25-37 doi 10.1007/s11010-009-0267-2. 52. Li F, Sethi G. Targeting transcription factor NF-kappaB to overcome chemoresistance and radioresistance in cancer therapy. Biochim Biophys Acta 2010;1805(2):167-80 doi 10.1016/j.bbcan.2010.01.002 S0304- 419X(10)00003-X [pii].

53. Debatin KM, Krammer PH. Death receptors in chemotherapy and cancer. Oncogene 2004;23(16):2950-66 doi 10.1038/sj.onc.1207558 1207558 [pii]. 54. Baritaki S, Huerta-Yepez S, Sakai T, Spandidos DA, Bonavida B. Chemotherapeutic drugs sensitize cancer cells to TRAIL-mediated apoptosis: up-regulation of DR5 and inhibition of Yin Yang 1. Molecular cancer therapeutics 2007;6(4):1387-99 doi 6/4/1387 [pii] 10.1158/1535- 7163.MCT-06-0521. 55. Dimberg LY, Anderson CK, Camidge R, Behbakht K, Thorburn A, Ford HL. On the TRAIL to successful cancer therapy? Predicting and counteracting resistance against TRAIL-based therapeutics. Oncogene 2013;32(11):1341-50 doi 10.1038/onc.2012.164 onc2012164 [pii].

Figure legends:

Figure 1. Schematic of Bayesian network Reverse-Engineering and Forward-

Simulation. A. Drug treatment (as an icon of drug pill), gene expression (blue hexagon)

and phenotypic data (orange square) are prepared for modeling via procedures outlined in

the supplementary section. B. Likely fragments for network reconstruction were identified

by scoring all 2- and 3-variable combinations with the constraint that drug treatment is

upstream of all other nodes. C. Parallel global network sampling results in an ensemble of

1024 network structures that explain the data. The probabilistic directionality computed

by the Bayesian framework allows inferences to be made about what lies upstream and

downstream of a particular variable of interest. D. Diversity of network structures captures

uncertainty in the model. Hypotheses can be extracted from the network ensemble by

completing Monte Carlo simulations of “what-if” scenarios. Down-regulating a transcript

indicated by red arrow would be expected to impact the phenotype (the orange square next

to dark blue arrow).

Figure 2. Model validation: A. Cell cycle analysis of MDA-MB-231 cells treated with

siRNA against 17 genes and NC siRNA. Cells were grown in DMEM-10%FBS, treated

with designated siRNA pools and synchronized in G1 with mimosine. Cells were released

from G1-block and cell cycle distribution was assessed by FACS at 10h post release. Data

are shown as mean +/- standard deviation (sd) for three independent experiments.

Statistical significance was determined by student t-test; * - p<0.05, ** - p<0.005, *** -

p<0.0005; “+” designates genes predicted by the model to affect G1-S transition. B-C.

Molecular inhibition of G1-S progression by KD. Cells were treated as in (a) and protein

and phospho-protein levels were measured by immunoblotting (IB); n/s – non-specific

band. Bands were normalized by total protein and plotted as fraction of NC-siRNA

treatment.

Figure 3. TRIB1 mediates expression of CCND1 via regulation of NFκB and AP1. A.

Schematic representation of -1080-Cyclin D1 (D1) pGL3 basic luciferase reporter

constructs and deletion mutants. AP1, EtsA/B (both Ets A and B were deleted), and both

NFκB sites (κB1 and κB2) were eliminated as shown. B. Cells were treated with NC or

TRIB1 siRNA pools for 24h followed by transfection with 1μg of CCND1-WT, AP1-

mutant, EtsA/B-mutant, or κB1/2-mutant promoter-reporter constructs. Luciferase activity

was measured and standardized by co-transfection with SEAP expression vectors. Results

represent the means and standard deviations from three independent experiments. (RLU,

relative light units). C. TRIB1 regulates NFκB-responsive promoter activity. Cells were

treated as in (B) and transfected with NFκB-responsive reporter vector and SEAP

expression control vector, and treated with TNFα for 4 h. Cells were co-transfected with

1μg vector-EGFP or TRIB1-EGFP and with NFκB -promoter vector/SEAP and stimulated

with TNFα for 4 h. D. TRIB1 knockdown leads to downregulation of NFκB target genes.

Levels of CXCL1, CSF2, and IL8 proteins were measured using Luminex assay for

secreted cytokines. E. TRIB1 knockdown inhibits NFκB signaling via activation of IκBa

and inhibition of p50 and pIKKα. IB analysis of cells treated with NC or TRIB1 siRNA

pools with indicated antibodies.

Figure 4. TRIB1 knockdown induces apoptosis and sensitizes cells to TRAIL-induced

apoptosis in breast cancer cells. A. TRIB1 knockdown inhibits ERK1/2 and AKT

phosphorylation and activation of downstream signaling. IB analysis of cells treated with

siRNA pools against NC or TRIB1 with indicated antibodies. B. MDA-MB-231, HS578T,

SUM159, MDA-MB-468, MDA-MB-436, and T47D cells were transfected with NC (grey

bars) or TRIB1 (black bars) siRNA pools, cells were stained with Annexin-V/PI and

analyzed by FACS. Data shown as mean+/- sd of three independent experiments. C.

TRIB1 knockdown sensitizes cells to TRAIL at non-lethal concentrations. Cells were

treated as in (B) and followed by stimulation with TRAIL and analyzed as in (B). D.

TRAIL–induced apoptosis occurs through inactivation of BCL2 and cleavage of BID. IB

analysis of cells treated NC or TRIB1 siRNA -/+ TRAIL with indicated antibodies. E.

Changes in YY1 and DR5 mRNA levels in response to TRIB1 siRNA. F. TRIB1

knockdown results in upregulation of DR5 protein.

Figure 5. Expression and survival analysis in primary tumors. A. Heatmap illustrating

Pearson’s correlation between the expression levels of a panel of 38 TRIB1, TRAIL, and

NFkB associated genes in a cohort of primary breast tumors (n=1980). Genes that were

significantly (p<0.05) associated with TRIB1 expression are indicated in the

accompanying table with their respective correlation coefficients. B. Kaplan-Meier and

Hazard Ratio plots and log-rank p-values comparing breast cancer specific survival

(BCSS) and overall survival (OS) distributions between individuals with TRIB1

amplification in the upper (66.66%) versus lower (33.33%) tertiles.

Bayesian network inference modeling identifies TRIB1 as a novel regulator of cell cycle progression and survival in cancer cells.

Rina Gendelman, Heming Xing, Olga K Mirzoeva, et al.

Cancer Res Published OnlineFirst January 13, 2017.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-16-0512

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2017/01/13/0008-5472.CAN-16-0512.DC1

Author Author manuscripts have been peer reviewed and accepted for publication but have not yet been Manuscript edited.

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/early/2017/01/13/0008-5472.CAN-16-0512. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.